Centrala begrepp
AcME-AD provides efficient and transparent explanations for anomaly detection models, enhancing trust and usability.
Sammanfattning
Anomaly Detection (AD) is crucial for various applications.
Traditional methods lack transparency, hindering trust.
AcME-AD offers model-agnostic interpretability for tabular data.
Sub-scores like Delta, Ratio, Change, and Distance-to-change aid in feature importance assessment.
Experimental results show AcME-AD's effectiveness on synthetic and real datasets.
Comparison with KernelSHAP and LocalDIFFI highlights AcME-AD's advantages.
Feature selection analysis demonstrates the relevance of AcME-AD's identified features.
Statistik
AcME-AD는 효율적이고 투명한 이상 탐지 모델 해석을 제공합니다.
전통적인 방법은 신뢰성을 저해하는 투명성이 부족합니다.
AcME-AD는 표 형식 데이터에 대한 모델에 중립적 해석을 제공합니다.
Citat
"AcME-AD offers local feature importance scores and a what-if analysis tool."
"The lack of transparency compromises the reliability of traditional Anomaly Detection methods."